Skip to content

Inconsistency between max_length and vllm_max_model_len in official examples #3956

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
kyrie-maker opened this issue Apr 22, 2025 · 1 comment

Comments

@kyrie-maker
Copy link

kyrie-maker commented Apr 22, 2025

Description

I noticed an inconsistency in the official examples in https://swift.readthedocs.io/en/latest/Instruction/GRPO.html regarding the relationship between max_length and vllm_max_model_len parameters.

In the official examples (both async mode and colocate mode), these parameters are set differently:

--vllm_max_model_len 8192 \
--max_length 4096 \

Questions

  1. Is this inconsistency intentional? If so, what's the rationale behind having different values for these parameters?
  2. What are the potential impacts of having different values for these parameters?

Expected Behavior

I would expect these parameters to be consistent since they both relate to sequence length handling. Looking forward to clarification on these parameters.

@hjh0119
Copy link
Collaborator

hjh0119 commented Apr 22, 2025

max_length refers to the length of the query (prompt) part, while vllm_max_model_len corresponds to the maximum length of the query + response.

@hjh0119 hjh0119 closed this as completed Apr 27, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants